Instructions to use Emreuludasdemir/teknofest2026-task3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LightGlue
How to use Emreuludasdemir/teknofest2026-task3 with LightGlue:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
File size: 16,528 Bytes
699f3cd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 | from __future__ import annotations
import json
from pathlib import Path
from statistics import mean
from typing import Iterable
from src.config.settings import MvpRuntimeSettings
from src.core.frame_state import DecodedFrame, FrameEnvelope
from src.core.video_io import iter_video_frames
from src.task3.matcher import Task3Matcher
from src.task3.no_match_logic import filter_no_match_candidates
from src.task3.reference_cache import ReferenceCache
from src.task3.verifier import verify_matches
from src.tools.report_paths import GENERATED_REPORTS_ROOT
def discover_task3_videos(reference_dir: str | Path | None = None) -> list[Path]:
base = Path(reference_dir) if reference_dir is not None else MvpRuntimeSettings().task3_eval_reference_dir
parent = base.parent if base.is_dir() else base
deduped: dict[str, Path] = {}
for candidate in list(parent.glob("*.MP4")) + list(parent.glob("*.mp4")):
deduped[str(candidate.resolve()).lower()] = candidate
return sorted(deduped.values())
def evaluate_task3_frames(
frames: Iterable[DecodedFrame],
*,
runtime_settings: MvpRuntimeSettings | None = None,
reference_dir: str | Path | None = None,
video_name: str = "task3_replay",
mode: str = "orb_template",
) -> dict[str, object]:
settings = runtime_settings or MvpRuntimeSettings()
cache = ReferenceCache()
cache.preload_from_directory(reference_dir or settings.task3_eval_reference_dir, orb_features=settings.task3_orb_features)
matcher = Task3Matcher(reference_cache=cache, runtime_settings=settings)
reference_ids = cache.list_ids()
total_frames = 0
raw_candidate_frames = 0
accepted_match_count = 0
rejected_verification_count = 0
ambiguity_suppression_count = 0
no_match_frames = 0
descriptor_path_count = 0
template_path_count = 0
false_positive_proxy_count = 0
match_scores: list[float] = []
total_candidates_generated = 0
total_candidates_accepted = 0
total_candidates_rejected = 0
total_candidates_rejected_by_gate = 0
total_candidates_rejected_by_score_filter = 0
yoloe_inference_ms_values: list[float] = []
lightglue_verify_ms_values: list[float] = []
homography_compute_ms_values: list[float] = []
effective_mode_counts: dict[str, int] = {}
fallback_reason_counts: dict[str, int] = {}
for frame_index, decoded in enumerate(frames):
total_frames += 1
frame = FrameEnvelope(
frame_url=f"http://task3-eval/frames/{frame_index + 1}/",
image_url=f"/task3/{frame_index + 1}.jpg",
video_name=video_name,
translation_x=0.0,
translation_y=0.0,
translation_z=0.0,
health_status="1",
metadata={"frame_index": decoded.frame_index, "image_width": decoded.width, "image_height": decoded.height},
)
raw_matches = matcher.match(frame, b"", reference_ids, decoded_frame=decoded, mode=mode)
task3_info = dict(matcher.last_run_info)
effective_mode = str(task3_info.get("effective_mode", mode))
effective_mode_counts[effective_mode] = effective_mode_counts.get(effective_mode, 0) + 1
fallback_reason = task3_info.get("fallback_reason")
if fallback_reason:
reason_text = str(fallback_reason)
fallback_reason_counts[reason_text] = fallback_reason_counts.get(reason_text, 0) + 1
generated_count = int(task3_info.get("candidates_generated", len(raw_matches)))
total_candidates_generated += generated_count
gate_rejected_count = int(task3_info.get("candidates_rejected_by_gate", max(generated_count - len(raw_matches), 0)))
total_candidates_rejected_by_gate += gate_rejected_count
yoloe_inference_ms_values.append(float(task3_info.get("yoloe_inference_ms", 0.0)))
lightglue_verify_ms_values.append(float(task3_info.get("lightglue_verify_ms_total", 0.0)))
homography_compute_ms_values.append(float(task3_info.get("homography_compute_ms_total", 0.0)))
if raw_matches:
raw_candidate_frames += 1
filtered = filter_no_match_candidates(
raw_matches,
min_score=settings.task3_min_score,
mode=mode,
yoloe_min_score=settings.task3_yoloe_min_score,
modality=decoded.modality,
yoloe_thermal_min_score=settings.task3_yoloe_thermal_min_score,
ambiguity_margin=settings.task3_ambiguity_margin,
suppression_mode=cache.get_candidate_suppression_mode(),
)
score_filter_rejected_count = max(len(raw_matches) - len(filtered), 0)
total_candidates_rejected_by_score_filter += score_filter_rejected_count
if raw_matches and not filtered and len(raw_matches) > 1:
ambiguity_suppression_count += 1
verified = verify_matches(
frame,
filtered,
decoded_frame=decoded,
min_inliers=settings.task3_match_min_inliers,
)
rejected_verification_count += max(len(filtered) - len(verified), 0)
final_accepted_count = len(verified)
total_candidates_accepted += final_accepted_count
total_candidates_rejected += max(generated_count - final_accepted_count, 0)
if not verified:
no_match_frames += 1
continue
for match in verified:
accepted_match_count += 1
score = float(match.metadata.get("match_score", 0.0))
match_scores.append(score)
source = str(match.metadata.get("matcher_source", ""))
if "template" in source:
template_path_count += 1
if score < 0.88:
false_positive_proxy_count += 1
elif "yoloe" in source:
descriptor_path_count += 1
yoloe_info = match.metadata.get("task3_yoloe", {})
if not bool(yoloe_info.get("verify_passed", False)):
false_positive_proxy_count += 1
elif "learned" in source:
descriptor_path_count += 1
if float(match.metadata.get("similarity", 0.0)) < settings.task3_learned_min_similarity:
false_positive_proxy_count += 1
else:
descriptor_path_count += 1
if float(match.metadata.get("inlier_ratio", 0.0)) < 0.45:
false_positive_proxy_count += 1
decision = _decide_learned_descriptor_need(
accepted_match_count=accepted_match_count,
false_positive_proxy_count=false_positive_proxy_count,
descriptor_path_count=descriptor_path_count,
template_path_count=template_path_count,
)
return {
"video_name": video_name,
"status": "ok",
"reference_count": len(reference_ids),
"total_frames": total_frames,
"raw_candidate_frames": raw_candidate_frames,
"accepted_match_count": accepted_match_count,
"rejected_verification_count": rejected_verification_count,
"ambiguity_suppression_count": ambiguity_suppression_count,
"no_match_suppression_rate": round(no_match_frames / max(total_frames, 1), 6),
"false_positive_proxy_count": false_positive_proxy_count,
"descriptor_path_count": descriptor_path_count,
"template_path_count": template_path_count,
"mean_match_score": round(mean(match_scores), 6) if match_scores else 0.0,
"effective_mode_counts": effective_mode_counts,
"fallback_reason_counts": fallback_reason_counts,
"fallback_reason": _summarize_reason_counts(fallback_reason_counts),
"fallback_active": bool(fallback_reason_counts),
"candidates_generated": total_candidates_generated,
"candidates_accepted": total_candidates_accepted,
"candidates_rejected": total_candidates_rejected,
"candidates_rejected_by_gate": total_candidates_rejected_by_gate,
"candidates_rejected_by_score_filter": total_candidates_rejected_by_score_filter,
"candidate_rejected_ratio": round(total_candidates_rejected / max(total_candidates_generated, 1), 6),
"yoloe_inference_ms_per_frame_avg": round(_safe_mean(yoloe_inference_ms_values), 6),
"lightglue_verify_ms_total_per_frame_avg": round(_safe_mean(lightglue_verify_ms_values), 6),
"homography_compute_ms_per_frame_avg": round(_safe_mean(homography_compute_ms_values), 6),
"decision": decision,
"mode": mode,
}
def evaluate_task3_baseline(
*,
runtime_settings: MvpRuntimeSettings | None = None,
output_dir: str | Path = GENERATED_REPORTS_ROOT,
mode: str = "orb_template",
) -> dict[str, object]:
settings = runtime_settings or MvpRuntimeSettings()
output_path = Path(output_dir)
output_path.mkdir(parents=True, exist_ok=True)
results: list[dict[str, object]] = []
for video_path in discover_task3_videos(settings.task3_eval_reference_dir):
try:
frames = iter_video_frames(
video_path,
frame_stride=settings.task3_eval_frame_stride,
limit=settings.task3_eval_frame_limit,
video_name=video_path.stem,
)
summary = evaluate_task3_frames(
frames,
runtime_settings=settings,
reference_dir=settings.task3_eval_reference_dir,
video_name=video_path.stem,
mode=mode,
)
except Exception as exc:
summary = {"video_name": video_path.stem, "status": "failed", "error": str(exc), "total_frames": 0}
results.append(summary)
aggregate = {
"video_count": len(results),
"ok_count": sum(1 for item in results if item.get("status") == "ok"),
"no_match_suppression_rate": round(_safe_mean(item.get("no_match_suppression_rate", 0.0) for item in results if item.get("status") == "ok"), 6),
"false_positive_proxy_count": int(sum(int(item.get("false_positive_proxy_count", 0)) for item in results if item.get("status") == "ok")),
"accepted_match_count": int(sum(int(item.get("accepted_match_count", 0)) for item in results if item.get("status") == "ok")),
"decision": _aggregate_decision(results),
}
payload = {"results": results, "aggregate": aggregate, "mode": mode}
if mode == "orb_template":
(output_path / "task3_baseline_summary.json").write_text(json.dumps(payload, indent=2), encoding="utf-8")
(output_path / "task3_baseline_table.md").write_text(render_task3_table(results), encoding="utf-8")
suffix_map = {
"orb_template": "orb",
"learned_descriptor": "learned",
"yoloe_vp_lightglue": "yoloe",
}
suffix = suffix_map.get(mode, mode.replace("-", "_"))
(output_path / f"task3_baseline_{suffix}_summary.json").write_text(json.dumps(payload, indent=2), encoding="utf-8")
(output_path / f"task3_baseline_{suffix}_table.md").write_text(render_task3_table(results), encoding="utf-8")
write_task3_comparison(output_path)
return payload
def write_task3_comparison(output_dir: str | Path) -> dict[str, object] | None:
output_path = Path(output_dir)
orb_path = output_path / "task3_baseline_orb_summary.json"
learned_path = output_path / "task3_baseline_learned_summary.json"
if not orb_path.exists() or not learned_path.exists():
return None
orb_payload = json.loads(orb_path.read_text(encoding="utf-8"))
learned_payload = json.loads(learned_path.read_text(encoding="utf-8"))
orb_aggregate = orb_payload.get("aggregate", {})
learned_aggregate = learned_payload.get("aggregate", {})
comparison_decision = decide_learned_descriptor_gain(orb_aggregate, learned_aggregate)
markdown = (
"| Metric | ORB/Template | Learned | Delta |\n"
"| --- | --- | --- | --- |\n"
f"| Accepted Match Count | {orb_aggregate.get('accepted_match_count', '-')} | {learned_aggregate.get('accepted_match_count', '-')} | "
f"{int(learned_aggregate.get('accepted_match_count', 0)) - int(orb_aggregate.get('accepted_match_count', 0))} |\n"
f"| False Positive Proxy | {orb_aggregate.get('false_positive_proxy_count', '-')} | {learned_aggregate.get('false_positive_proxy_count', '-')} | "
f"{int(learned_aggregate.get('false_positive_proxy_count', 0)) - int(orb_aggregate.get('false_positive_proxy_count', 0))} |\n"
f"| No-match Suppression Rate | {orb_aggregate.get('no_match_suppression_rate', '-')} | {learned_aggregate.get('no_match_suppression_rate', '-')} | "
f"{round(float(learned_aggregate.get('no_match_suppression_rate', 0.0)) - float(orb_aggregate.get('no_match_suppression_rate', 0.0)), 6)} |\n"
f"| Decision | {orb_aggregate.get('decision', '-')} | {learned_aggregate.get('decision', '-')} | - |\n"
f"| Faz 6 Learned Gain | - | {comparison_decision} | - |\n"
)
(output_path / "task3_baseline_comparison.md").write_text(markdown, encoding="utf-8")
return {"orb": orb_aggregate, "learned": learned_aggregate, "decision": comparison_decision}
def render_task3_table(results: list[dict[str, object]]) -> str:
lines = [
"| Video | Status | Frames | Accepted | Rejected | No-match Rate | FP Proxy | Descriptor | Template | Fallback | Decision |",
"| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |",
]
for item in results:
lines.append(
"| {video} | {status} | {frames} | {accepted} | {rejected} | {no_match} | {fp} | {descriptor} | {template} | {fallback} | {decision} |".format(
video=item.get("video_name"),
status=item.get("status"),
frames=item.get("total_frames", 0),
accepted=item.get("accepted_match_count", "-"),
rejected=item.get("rejected_verification_count", "-"),
no_match=item.get("no_match_suppression_rate", "-"),
fp=item.get("false_positive_proxy_count", "-"),
descriptor=item.get("descriptor_path_count", "-"),
template=item.get("template_path_count", "-"),
fallback=item.get("fallback_reason", "-"),
decision=item.get("decision", "-"),
)
)
return "\n".join(lines) + "\n"
def _decide_learned_descriptor_need(
*,
accepted_match_count: int,
false_positive_proxy_count: int,
descriptor_path_count: int,
template_path_count: int,
) -> str:
if accepted_match_count == 0:
return "belirsiz"
if false_positive_proxy_count > max(2, accepted_match_count // 5):
return "gerekli"
if descriptor_path_count >= template_path_count and false_positive_proxy_count == 0:
return "henuz_gereksiz"
return "belirsiz"
def _aggregate_decision(results: list[dict[str, object]]) -> str:
decisions = [str(item.get("decision", "belirsiz")) for item in results if item.get("status") == "ok"]
if not decisions:
return "belirsiz"
if any(item == "gerekli" for item in decisions):
return "gerekli"
if all(item == "henuz_gereksiz" for item in decisions):
return "henuz_gereksiz"
return "belirsiz"
def decide_learned_descriptor_gain(orb_aggregate: dict[str, object], learned_aggregate: dict[str, object]) -> str:
orb_fp = int(orb_aggregate.get("false_positive_proxy_count", 0))
learned_fp = int(learned_aggregate.get("false_positive_proxy_count", 0))
orb_no_match = float(orb_aggregate.get("no_match_suppression_rate", 0.0))
learned_no_match = float(learned_aggregate.get("no_match_suppression_rate", 0.0))
orb_accept = int(orb_aggregate.get("accepted_match_count", 0))
learned_accept = int(learned_aggregate.get("accepted_match_count", 0))
fp_improved = learned_fp <= int(round(orb_fp * 0.8))
no_match_ok = (learned_no_match - orb_no_match) <= 0.10
accept_ok = learned_accept >= int(round(orb_accept * 0.85))
if fp_improved and no_match_ok and accept_ok:
return "kazanc_var"
if learned_fp >= orb_fp and learned_accept <= orb_accept:
return "kazanc_yok"
return "belirsiz"
def _safe_mean(values: Iterable[float]) -> float:
filtered = [float(value) for value in values]
return mean(filtered) if filtered else 0.0
def _summarize_reason_counts(counts: dict[str, int]) -> str | None:
if not counts:
return None
if len(counts) == 1:
return next(iter(counts))
return "mixed"
|